| 1 | #TODO: setRefClass... to avoid copy data !! |
| 2 | #http://stackoverflow.com/questions/2603184/r-pass-by-reference |
| 3 | |
| 4 | #fields: data (can be NULL or provided by user), coeffs (will be computed |
| 5 | #con can be a character string naming a file; see readLines() |
| 6 | #data can be in DB format, on one column : TODO: guess (from header, or col. length...) |
| 7 | |
| 8 | |
| 9 | writeTmp(curves [uncompressed coeffs, limited number - nbSeriesPerChunk], last=FALSE) #if last=TRUE, close the conn |
| 10 | readTmp(..., from index, n curves) #careful: connection must remain open |
| 11 | #TODO: write read/write tmp reference ( on file in .tmp/ folder ... ) |
| 12 | |
| 13 | epclust = function(data=NULL, K, nbPerChunk, ..., writeTmp=ref_writeTmp, readTmp=ref_readTmp) #where to put/retrieve intermediate results; if not provided, use file on disk |
| 14 | { |
| 15 | |
| 16 | |
| 17 | #on input: can be data or con; data handled by writing it to file (ascii or bin ?!), |
| 18 | #data: con or matrix or DB |
| 19 | |
| 20 | #1) acquire data (process curves, get as coeffs) |
| 21 | if (is.numeric(data)) |
| 22 | { |
| 23 | #full data matrix |
| 24 | index = 1 |
| 25 | n = nrow(data) |
| 26 | while (index < n) |
| 27 | { |
| 28 | writeTmp( getCoeffs(data) ) |
| 29 | index = index + nbSeriesPerChunk |
| 30 | } |
| 31 | } else if (is.function(data)) |
| 32 | { |
| 33 | #custom user function to retrieve next n curves, probably to read from DB |
| 34 | writeTmp( getCoeffs( data(nbPerChunk) ) ) |
| 35 | } else |
| 36 | { |
| 37 | #incremental connection |
| 38 | #read it one by one and get coeffs until nbSeriesPerChunk |
| 39 | #then launch a clustering task............ |
| 40 | ascii_lines = readLines(data, nbSeriesPerChunk) |
| 41 | seriesChunkFile = ".tmp/seriesChunk" #TODO: find a better way |
| 42 | writeLines(ascii_lines, seriesChunkFile) |
| 43 | writeTmp( getCoeffs( read.csv(seriesChunkFile) ) ) |
| 44 | } else |
| 45 | stop("Unrecognizable 'data' argument (must be numeric, functional or connection)") |
| 46 | |
| 47 | #2) process coeffs (by nbSeriesPerChunk) and cluster in parallel (just launch async task, wait for them to complete, and re-do if necessary) |
| 48 | |
| 49 | |
| 50 | #3) apply stage 2 (in parallel ? inside task 2) ?) |
| 51 | } |
| 52 | |
| 53 | getCoeffs = function(series) |
| 54 | { |
| 55 | #... return wavelets coeffs : compute in parallel ! |
| 56 | } |